No waitlist.
No approval.
No catch. We looked.

We have been heads down, not asleep.

You signed up and then heard very little from us. We would love to blame the algorithm. It was on purpose: Kaitoi is built by a very small team, and every week spent talking about the product is a week not spent building it. So we built. Design partners ran real projects through Kaitoi, hit the walls before you could, and described the experience in detail we did not always enjoy reading. What came out of that is not a roadmap. It is the deeply unglamorous work of making a very wide product hold together under real use.

We are still a very small team. The upside: when you report something, it lands on the desk of the person who wrote the bug, usually the same day. The downside: that person is also doing four other things, and we would rather say so than imply we have a department.

Kaitoi is now in open beta, and anyone can sign up.

We could open with a feature tour. Instead, here are the two things worth knowing before anything else. The tour is right behind them, and fair warning, it goes on a bit.

Privacy

Your work is not our training data.

We never train on your work. We never sell it. We never share it beyond what running the job you asked for actually requires. That is the whole position, not the friendly summary of a longer document that says the opposite.

To be precise rather than merely reassuring: your projects and outputs are stored in Kaitoi, and when a node calls a third-party provider your prompt and your assets go to that provider under their terms, which we do not control and will not pretend to. Your own key changes whose account it runs under, not whether it happens. We tell you which provider handled which run, every time. All of this is in the privacy policy, which we wrote in actual words.

Lab

A moodboard you can talk to.

Not every project starts with a graph. Some start with a half-formed idea and a blank wall you have been staring at since Tuesday.

Lab is conversation on one side and a free-form canvas on the other. Describe what you are after and images, video, audio, and 3D land on the board as you go. Mention any item or group by name to refine it, combine it, or spin variations. The whole board stays with the project, bad ideas included, which is how you remember why they were bad.

You can also teach it your own moves. Turn any node or template into a skill, and your pipeline becomes something Penny can reach for mid-conversation.

And Penny is not just forwarding your prompt to a model and hoping for the best.

In advertising it works with Ad Muse, our own creative library. Give it a brief and it pulls real campaign references, works out why they landed, and comes back with original concepts instead of a wall of things that look like other things.

Pick the ones worth chasing and it develops them: scripts, copy decks, art direction, or a full campaign with the hero film, the visual system, social, experiential, and print all hanging off one strategic thread. Then you build the actual assets in the same workspace, which beats emailing a deck to yourself at midnight.

Models and LLMs

A new model ships. One click, and it is a node.

More than 1,400 generative models are available, and image, video, audio, and 3D are only the headline categories. When a provider releases something new, you do not file a feature request and wait for us to notice it. You add it yourself, in one click, and it turns up as a working node.

The catalog keeps up with the industry because you never have to wait for us to keep up with it. We are calling that empowerment.

Always the latest version of each. And that is a sample. There are more models than that, more kinds of model than that, and then whatever shipped while you were reading this paragraph.

A number that size is only useful if you can find the right one, so the list is curated and searchable: models, nodes, templates, files, and outputs all in one place, sorted by what they actually do rather than by who shouted loudest on launch day.

Then the language models. Every frontier one, and about 400 more.

Those 1,400 are the models that make things. Language models are counted separately, and there are around 400 of them: the ones that write, reason, and run Penny.

OpenAI, Anthropic, and Google Gemini are supported directly. On top of that, full OpenRouter support puts every other frontier model within reach, along with hundreds of cheaper ones that are quietly excellent and never get a keynote.

Use them in your graphs, or pick which one runs the agent. Expensive when the task deserves it, cheap when it is renaming files.

MCP

MCP in both directions.

Add any MCP server to your workspace and Penny can use it, the same way it uses everything else.

More interesting: Kaitoi is an MCP server itself.

Connect Claude Code, Codex, or any other MCP client to your Kaitoi workspace and let it build nodes, run graphs, and pull results back into whatever you are working in. Your coding agent, quietly operating a media platform behind your back. Available on Indie and Studio.

Credits

Free to start, and you can earn more.

Your account starts with free credits, ready to claim. After that, the fastest way to get more is to make Kaitoi better:

We credit your account for all three. Yes, we will pay you to complain about us, provided you complain accurately.

Bugs, ideas, and half-finished work all live on Discord. Come be loud.

Questions? Reach us at support@kaitoi.io. A human reads it. Usually the one who broke it.